nsqd: add flag --min-output-buffer-timeout with default 25ms

[ploxiln]

Lower timeouts are more expensive for nsqd, so it seems odd to put
a maximum on the timeout a client can request, but not a minimum.

Also change --max-output-buffer-timeout default to 30 seconds,
up from 1 second. Again, it's cheaper/easier for nsqd.
But we may still want to help people avoid unwittingly setting
it close to (or above) the default --msg-timeout (60 seconds).

---

I wasn't familiar with output_buffer_timeout until #1125 came up. I don't have a real need for these options changes, they just seem to make a bit more sense. But what do you think @mreiferson ?

possibly useful links to refresh memory:
043b79a
https://nsq.io/clients/tcp_protocol_spec.html#identify

[ploxiln]

(The reason I was a bit confused about disabling the timeout is because I thought I read somewhere that -1 would disable output buffering, but that's just for output_buffer_size. So it's odd that there's a --max-output-buffer-timeout but if the client requests -1 for output_buffer_timeout it gets a sort of infinitely long timeout.)

[ploxiln]

This makes me think that if either output_buffer_size or output_buffer_timeout is disabled, the other should be too. I don't think that currently is the case.

[mreiferson]

Seems reasonable, LGTM.

[ploxiln]

I'm going to let it sit for just a bit, then re-review and manually test a bit, before merge.
(I opened this partly for discussion, and was expecting some direction :)

[mreiferson]

Ha, it seemed like the conversation (with yourself) sort of concluded 😜? The changes don't seem controversial, what do you feel like we need to discuss?

[ploxiln]

Maybe nothing :) Just maybe, for backwards compatibility, or some other practical reason, we should be careful changing some bit here.

Anyway I'll just do a sanity check (min/max is enforced) sometime in the next few days.

[ploxiln]

Did some local testing, looked good.

Side note: pynsq does not currently allow -1 for output_buffer_size or output_buffer_timeout